You are on page 1of 11

Expert Systems

with Applications
Expert Systems with Applications 32 (2007) 868–878
www.elsevier.com/locate/eswa

Design and implementation of a fuzzy inference system


for supporting customer requirements
a,c,*
Ying-Shen Juang , Shui-Shun Lin b, Hsing-Pei Kao c

a
Department of Business Administration, Chung-Hua University, Taiwan, ROC
b
Department of Business Administration, National Chinyi Institute of Technology, Taiwan, ROC
c
Institute of Industrial Management, National Central University, Taiwan, ROC

Abstract

Efficient and effective response to the requirements of customers is a major performance indicator. Failure to satisfy customer require-
ments implies operational weaknesses in a company. These weaknesses will damage both the rights of customers and the reputation of
the company. The traditional method of handling customer requirement for a machine tool manufacturer was dominated by manual
process and subjective decision. In this study, we improved the operation process of handling customer requirement. The framework
of a customer requirement information system (CRIS) for machine tool manufacturers was then analyzed, integrating rule-based fuzzy
inference and expert systems, and a prototype system developed. The CRIS supports both customers and service personnel in providing a
systematic way of fulfilling and analyzing customer requirements. The system was installed and operated in a machine tool manufacturer
and the performance was found promising.
Ó 2006 Elsevier Ltd. All rights reserved.

Keywords: Fuzzy inference; Expert systems; Customer requirements; Fault diagnosis; Integrated information system

1. Introduction suppliers, customers, and constituents. Increasing demand


and hands-on users are rendering the customary model of
The machine tool sector consists of more than 1400 business intelligence applications, originated within depart-
companies in Taiwan, with over 26,000 employees. This ments and isolated from the enterprise, inefficient and inef-
sector thus ranks fifth worldwide as far as manufacturing fective. Information technologies are anticipated to
and export are concerned (Perng, Tsai, & Lin, 2003). The transform this paradigm, introducing information-rich
Taiwanese government thus prioritized this sector as one interactive capabilities to the e-business environment. Fur-
of 10 major industries for intensive development in the first thermore, globalization and information technology affect
decade of the new millennium. The machine tool sector industries radically. Many companies have become increas-
faces extreme pressure from global competition and infor- ingly aware of the significance of managing customer rela-
mation technology change. The major pressure arises from tionships. Concepts such as customer satisfaction,
the need for broad delivery of information to everyone who customer-orientation and service differentiation have
affects business processes with rapid time-to-market and become critical initiatives in developing the competitive
low cost-of-ownership. To confront this challenge, the advantages of a company.
industry requires business intelligence, not just for a select However, the findings in Kalakota and Robinson (1999)
few, but for everyone—employees, managers, partners, show, developing a new customer costs more than six times
the effort of keeping a present one. Hence, any demand of
the existing customer needs to be treated more positively.
*
Corresponding author. Tel.: +886 3 5186060; fax: +886 3 5186570. Customer requirement management seeks to ascertain cus-
E-mail address: ysjuang@chu.edu.tw (Y.-S. Juang). tomer needs and solve or fulfill them. Accelerating response

0957-4174/$ - see front matter Ó 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2006.01.052
Y.-S. Juang et al. / Expert Systems with Applications 32 (2007) 868–878 869

depends on a more advanced information system and the effectively at a local manufacturer. More specifically, the
exploitation of a specific method. objectives were as follows: (1) using fuzzy inference to sur-
In manufacturing industries, cases of utilizing informa- mount ambiguous verbal description of requirements from
tion technology to promote performance are growing sig- a customer; (2) analyzing the information demands of
nificantly. An example of this is the after-sale service CRIS and sequencing the integrated system framework
management information system for the machine tool based on fuzzy inference and expert systems; and (3) con-
industry proposed by Tsai, Shuei, Perng, Lin, and Yao structing a prototype of CRIS for the machine tool
(2001). The system handles data generated from merchan- industry.
dise service, such as repair request and maintenance his-
tory. It is a management information system for 2. Literature review
customer requirement. However, their system is incapable
of handling the requirements expressed by customer with The behavior of consumer requirements has been defined
vague semantics. in many studies (Day, 1980; Hart, Heskett, & Sasser, 1990;
For most situations in machine tool manufactures, the Heskett, Jones, Loveman, Sasser, & Schlesinger, 1994;
customer requirements are handled by service personnel Jacoby & Jaccard, 1981; Joentgen, Mikenina, Weber, Zeug-
manually in different department with their subjective rec- ner, & Zimmermann, 1999; Lewis, 1982). There are two
ognition and responses. However, this appears tedious basic concepts: (1) The behavior of consumer demands is
when there are diverse customer needs and inconsistent caused by dissatisfaction of the consumers. (2) The response
departmental policy. For instance, due to insufficient tech- to the behavior of consumer demands divides into two
nical knowledge and lack of awareness about machine spec- types: behavioral response and non-behavioral response.
ifications, customers may present requests with some Behavioral responses include conveying the negative images
inexact meanings. Our interview to major machine tool ser- to his/her friends and complaining directly to retailers, or to
vice managers also revealed that the most significant prob- the third parties, such as the Consumer Protection Founda-
lems the companies encountered, i.e., customer requirement tion. A non-behavioral response forgets the unsatisfactory
frequently received with verbally vague description and the experience without taking any further actions. Customer
customer service is not performed systematically. We requirements management is a procedure adopted to solve
noticed and recognized the practical difficulties in handling customer requirements and establish customer confidence
customer requirement, and thus developed a systematic in a company. The industry uses requirement data to alter
means of handling them. The detailed analysis was shown product, correct service weaknesses and thus enhance com-
in the sections of demand analysis and system analysis. pany reputation. Requirement data enables employees to
Fuzzy inference and expert systems were utilized in this immediately comprehend the customer requirements as
study to deal with events with fuzzy demands. When cus- problems are encountered. Because customer requirements
tomer requirements are encountered, fuzzy inference is are handled after the product or service is sold, require-
employed to clarify the unclear wordings. Expert systems ments management is also called service remedy. Gronroos
are then used to infer the root causes of the problems, sug- (1988) notes that the tactics of service remedy involve mea-
gesting suitable solutions to those problems. Furthermore, sures adopted when the supplier service fails.
this system also helps the customer service department to Cristo (1997) discovered that the extent of customer sat-
assess customer feedback, integrating it with after-sale ser- isfaction is inversely proportional to the response time. The
vice records. By analyzing the service records, product qual- longer the customer waiting time is, the more significant
ity can be improved and the service improved. the inverse relationship. When customer requirements
Additionally, the future electronic trading environment occur, appropriate and quick requirement management is
will center on business networks or community-like busi- of importance.
ness models. Alternatively, information technology enables In related studies for the machine tool sector, Perng, Li,
digitalization of the supply chain. The significance of digita- Tsai, and Lin (1996) and Tsai et al. (2001) targeted mainte-
lization resides in synergy resulting from sophistication of nance service systems. Their studies proposed some related
external trade environments rather than any immediate frameworks, but the indistinct features of customer
benefit from computerization. Conversely, applying digita- requirements were not discussed. Perng et al. (1996)
lization when the environment matures may already be too claimed that the difficulty of maintenance service in prac-
late (Tsai, Lu, Perng, Lin, & Chu, 2001). Therefore, the tice encompasses: (1) training effect not significant, (2)
motivation to apply information technology to customer repair process and material unrecorded in maintenance
requirement management, and thus to develop an inte- report and (3) component or material for early generation
grated customer requirement information system (CRIS) product not managed. Perng et al. (1996) also stated that
is to join the forces of all participants in the entire service building an information system for machine fault diagnosis
process, enhancing prompt customer requirement handling. raises many difficulties. First, due to the lack of engineering
This study analyzed and designed a CRIS for the knowledge, maintenance information was not accumulated
machine tool industry. The implemented system was there- in practice. Second, the technology revolution leads to a
after installed on an application server and functioned short product cycle, resulting in greater difficulties with
870 Y.-S. Juang et al. / Expert Systems with Applications 32 (2007) 868–878

repair tasks. The maintenance tasks mainly rely on skilled tool manufacturers and sales agents were interviewed. All
technicians which take a long period of time to train. Ou were managers working in the service departments of man-
(1997) discovered that the major customer complaint of ufacturers ranked within the top 10 for sales figures from
international machine tool buyers related to the lack of 1999 to 2003 in Taiwan.
quick response to their requests.
Zadeh, who first depicted fuzzy sets in 1965, advocated 3.1. The machine tool sector
the application of fuzzy set theory in quantitative measures
of the human thinking process (Zadeh, 1975). The practical The Taiwanese machine tool industry is one of the most
application system of fuzzy control was developed in Eur- potential industrial activities on the island. Machine tool
ope. Chi and Zhang (1999) stated that fuzzy semantics han- enterprises must have reliable operational ability to attract
dling was a widely applied methodology. Chen (2001) used international orders in the highly competitive global envi-
fuzzy set theory to implement a circuitry framework of a ronment. Five major characteristics of the machine tool
rhetoric fuzzy controller. industry were isolated from the interview materials:
Various applications of fuzzy inference can be found in
the literature (Berkan & Trubatch, 1997; Chen & Pham, 1. The machine tool industry has critically close relations
2000; Ruan & Kerre, 2000). However, an integrated infor- with the aerospace industry, electronics, the automobile
mation system for dealing with customer requirement is industry and defense. It plays an important supportive
not implemented. Javadpour and Knapp (2003) focused and cooperative role with other industries.
on the implementing a predictive neural network for use 2. The machine tool industry, producing high precision
as an operator’s aid in diagnosing faults with high predic- machining tools, is technology intensive.
tion accuracy in an automated manufacturing environ- 3. Demand for machine tools in market is dramatically
ment. To evaluate the performance of the model, the influenced by economical cycles.
network was assessed with both simulated time series and 4. The number of components and range of materials is
real time machine vibration data gathered in lab experi- large. Enterprises frequently encounter urgent requests
ments. However, the fuzzy neural network approach was for changes of design.
not applied to customer requirement. Joentgen et al. 5. While enterprises need to continuously invest on tech-
(1999) proposed a clustering-based dynamic method for nology and human resources, the returns on investment
early recognition of changes in a machine’s state and thus are relatively low.
for automatic fault detection. This recognition system
requires less expert knowledge than traditional approaches.
However, the collaborative approach ensuring efficient 3.2. Essential elements of the interview
information sharing was not employed.
Expert systems are computer programs developed In-depth interviews were conducted to build up an infor-
around the thinking processes of the expert and are capable mation system for dealing with customer requirements so
of answering complicated questions. Many successful that most of the problems mentioned earlier could be
instances utilizing expert systems are available to handle solved. Six major companies were selected for analysis.
customer requirement problems. For instance, Su (2000) The main analytical contents were as follows: (1) basic
proposed a malfunction recovery mechanism allowing information about the interviewed company, (2) the com-
maintenance personnel and decision makers to solve main- pany situation with regard to computer and networking
tenance tasks cooperatively. Lin (1998) used expert system usage, (3) the present process and condition of customer
to detect machine breakdowns. Perng et al. (1996) applied requirement handling and management and (4) the demand
neural network and expert systems to machine fault diag- for a CRIS for machine tool service.
nosis. However, although the operation processes of expert The interview conclusion was that the industry urgently
systems were thoroughly discussed and a prototype system needs an information system to handle customer require-
was constructed, requirements from customers still could ment and enhance customer relationships. The interview
not be handled. Liu and Chen (1995) developed a machine material also uncovered many key factors and concerns
troubleshooting expert system through a fuzzy multiattrib- for developing such a system. The most significant prob-
ute decision-making approach. This system consists of five lems the companies faced are the customer requirement fre-
components and improves the efficiency of the diagnostic quently received with verbally vague description and the
process. Fang (1995) proposed a method for on-line customer service are not performed systematically. The
machine condition monitoring involving a fuzzy feature- operational practice, system analysis and system develop-
state relationship matrices and expert system. ment in the study will now be discusses in detail.

3. Demand analysis 3.3. Practical operation process

To analyze the functional demand of a CRIS, and to The operational process was executed inefficiently in the
fully augment system usability, managers from six machine following manner shown in Fig. 1. First, customer require-
Y.-S. Juang et al. / Expert Systems with Applications 32 (2007) 868–878 871

Customer including service reports, customer response sheets and part


Requirement
request forms. Those sheets or forms were then sent to
the department manager for approval. If the requirements
Classification were made and fulfilled via phone or fax, the forms and
sheets were omitted, that is, no any information was
recorded. The process then went to billing, executed by
Maintenance Request General
the accounting department, for the service charge, man-
Requirement hour charge and/or maintenance part charge. The inventory
amount was also updated if the parts were checked out from
Fault Analysis Via
Phone
stock.
Send to Related
Departments
Can be Solved
4. System analysis and design
N
by Customer?
4.1. System analysis
Y

Phone or FAX The industry is not using CRISs to assist their service
Service tasks or handle their service information, nor for further
analysis of service information. The realization of the need
Request of Maintenance
Problems Y Parts or Maintenance of the industry stimulated the analysis and development of
Exist? Report a CRIS. Developing a CRIS is largely an attempt to enable
the sharing of enterprise resources, such as service informa-
Delivery
tion and manpower. With the aid of a CRIS, the customer
On-site
Service
N Parts service will be completed more efficiently. This will also
help release customers from their inconvenient situation,
consequently promoting customer royalty and increasing
Reporting the customer satisfaction. To achieve this goal, an enter-
Follow-up
N prise should provide a standard of operational procedure
for maintenance service personnel to work by. Because dif-
Manager
Approval ferent manufacturers usually develop very different pro-
cesses, a standard procedure for customer requirements
handling appears to be urgently needed. This investigation
Charging
Process
addresses the operational process, system analysis and sys-
tem design for a CRIS.

Case Filed Y 4.2. System flow

Fig. 1. Customer requirement handling process in practice. Fig. 2 illustrates the CRIS process. The system is divided
into two subsystems: general requirement handling and
ments were logged via phone, FAX or maintenance person- machine fault diagnosis. Customer problems first collected
nel. Basic customer information such as address, phone and classified. Service personnel directed the problem to
number and the machine model was recorded by either the relevant subsystem triggering the handling process.
the service person in the call center or maintenance person- For general requirements, the classification process is
nel. The requirement responses were then analyzed online selected, fuzzy semantic treatment is instituted if needed,
or sequentially called/faxed back to the customer. The and the problem is recorded in the requirement recording
problem-solving hints or instructions were sent to the cus- system.
tomer, and the problems were expected to be solved step by For machine fault related requirements, the machine
step by customers. fault diagnosis system will be activated if the ambiguity
If the customers could not solve the problem by them- checking of the verbal expression from a customer is
selves, the maintenance personnel offered two further forms cleared. If the verbal expression is ambiguous, the fuzzy
of service: (1) onsite service and (2) maintenance part deliv- handling process will be activated.
ery service. Before performing onsite service, service per- Following processing the general requirement or machine
sonnel needed to fill out a service request form. For those fault diagnosis, system stores the requirement treatment
requirements able to be met by delivering parts for self- details in the requirement recording system. Finally, the
replacement, the service personnel followed up the situation data in the requirement recording system are conveyed to
to ensure the requirement was fulfilled. After service was the requirement analysis system for further evaluation.
completed, either onsite or by part delivery, the service Some statistical computing was implemented in the proto-
personnel reported the process by filling out service records, type system, such as the defect rate, or requirement rate of
872 Y.-S. Juang et al. / Expert Systems with Applications 32 (2007) 868–878

Customer Requirement A0 CRIS


A1 Requirement Classification System

Classification A11 Human Related Complaint


A12 Machine Related Complaint
Machine Related
Request A2 Fault Diagnosis System
A21 Fuzzy Reasoning System
A22 Expert System
Ambiguous?
A23 Rulebase Management System
General
Y A3 Requirement Recording System
Request
A31 Customer Tracking System
Fuzzy
N Handling A32 Personnel Recording System
A33 Data Query System
Machine Fault A4 Requirement Analysis System
Diagnosis Module
A41 Breakdown Rating Analysis
Recording Module A42 Maintenance Supplies Analysis
A43 Machine Performance Rating
A44 Personnel Performance Rating
Analysis Module
Fig. 3. Framework of CRIS in IDEF0 structure.
Fig. 2. System flow.

level of an oil pump. Following the fuzzy inference mecha-


a certain customer in a particular period of time. Statistical nism, the output can be a fuzzy set or a precise set of cer-
computing can be extended to many aspects, and is support- tain features. Fuzzy inference infers the results from the
ive of decision making. existing knowledge base. For after-sale service personnel,
Based on the proposed operational process and the it is an important step to know about the customer require-
functional demand of a CRIS, a system analysis was con- ment thoroughly when solving requirement problems. They
ducted with IDEF0, which is a structured system analysis usually spend considerable time finding out the truth of a
tool. In the IDEF0 model, activities can be described by customer’s requirement, or the original customer intention.
their inputs, outputs, controls, and mechanisms (ICOM). To conquer this problem, a method of synthetic fuzzy eval-
In addition, the description of activities can be recursively uation was adopted to construct the inference mechanism
refined into greater and greater detail until the model is as in this study.
descriptive as necessary for implementation (Bravoco & Fig. 5 illustrates the framework of the fuzzy knowledge
Yadav, 1985; Mayor, Benjamin, Bruce, & Painter, 1995; base and the relationship between elements. The fuzzy
Perng et al., 2003; Ross, 1985). knowledge base comprises fuzzy concept base, fuzzy prop-
A system analysis result was derived for efficient, eco- osition base, fuzzy rule base and fuzzy strategy base.
nomical and customized establishment of a CRIS. The
result was further extended to the system design stage to 4.3.1. Fuzzy concept base
constructing a prototype illustrating the feasibility of the This contains the terminology and relevant predicate of
system. Figs. 3 and 4 are two examples indicating the sys- a verbal expression. Terminology is in the domain of the
tem processes depicted by IDEF0. The CRIS contains four fuzzy set, possesses many pre-defined dismemberment val-
subsystems, namely requirement classification system ues denoted by predicates.
(RCS), fault diagnosis system (FDS), requirement record-
ing system (RRS), and decision analysis system (DAS). 4.3.2. Fuzzy proposition base
Membership functions accrue to the fuzzy proposition,
4.3. Fuzzy inference which was induced from fuzzy concept base. There are
numerous types of membership functions, such as S-shape,
This system incorporates fuzzy inference and rule-based Z-shape, and P-shape, all easily definable with equations
expert systems. Fuzzy inference in this system refers to the and parameters. For example, if the general fuzzy set is
use of computer programs to execute inference work expressed as
resembling what humans do daily. Inputs and outputs
are two basic elements in a system using fuzzy handling A ¼ fðx; lA ðxÞÞg; x2X
approaches. The input constitutes some ambiguous verbal
semantics or unclear concepts for a specific event, such as where l denotes the membership function, and (x, lA(x)) is
the vague description of motor temperature or the leakage a singleton, then a fuzzifier given by
Y.-S. Juang et al. / Expert Systems with Applications 32 (2007) 868–878 873

After-Sale Service Document Customer Response and Satisfaction Level

Service
Requirement Item
Personnel ID Requirement
Classification Requirement Record
General Statement System Problem Solving Method
Request No.
A1
Machine ID
Part Used
Problem
Rule Base
Symptom Fault Diagnosis
Customer ID System Recommendation
Fault Statement
A2 Fault Number
Knowledge Base Root Cause
Inference Engine
Customer Info
Time Interval Requirement
Recording System
Technician Info
A3
Service Personnel Info Fault Record
Fault Allowance
Rate Decision Analysis General Problem Solution
Confidence Level
System
Machine Info Decision Analysis Info
A4

Service Personnel Info System Maintenance Technician

Fig. 4. System process depicted by IDEF0.

Fuzzy strategy base 4.3.4. Fuzzy strategy base


This contains the algorithms for computing the condi-
Algorithms for condition
and conclusion computing tion part and the conclusion part. A proposition might
encompass many conditions. An appropriate fitness of a
rule had to be found so that the conclusion can be drawn.
This is carried out by a process of implication. A member-
Fuzzy concept base Fuzzy proposition base Fuzzy rule base ship function that defines the implication relation can be
expressed in a number of ways. To illustrate the operation,
Terminology
we assume that we have the following simple conditional
Proposition Fuzzy rule
Predicate
proposition (canonical rule):
IF X is A THEN Y is B
Fig. 5. Framework of fuzzy knowledge base. The implication relation is defined by
Z
Rðx; yÞ ¼ lðx; yÞ=ðx; yÞ
x;y
1 where linguistic/fuzzy variable X and Y take the value of A
lðxÞ ¼ K ; x2X
1 þ ðx=K 2 Þ 1 and B, respectively, and l(x, y) is the membership function
of the implication relation. The membership function is
produces an S-shaped curvature. K1 and K2 are called the denoted by
exponential and denominational fuzzifiers, respectively.
lðx; yÞ ¼ lA ðxÞ ^ lB ðyÞ
By having controllable parameters such as K1 and K2,
adaptive fuzzy algorithms can be developed. The symbol ^ corresponds to intersection operation.

4.3.3. Fuzzy rule base 4.4. Numerical illustration


The fuzzy proposition is then presented in IF–THEN
format and constitutes the rule base. Specifically, a finite To demonstrate the operation of these elements, an
fuzzy logic implication statement in the rule base was example is given as follows:
described by a set of general fuzzy IF–THEN rules contain- The fuzzy concept base has a terminology of ‘‘oil leak-
ing only the fuzzy logical AND operation, in the form ‘‘IF age’’ with five terms of predicates, namely very few, few,
a11 is A11 AND . . . AND a1n is A1n THEN b1 is B1.’’ fair, much, very much. The membership function for each
874 Y.-S. Juang et al. / Expert Systems with Applications 32 (2007) 868–878

M (X) 4.5. Expert systems


1
Expert systems are capable of acquiring, analyzing, and
processing knowledge and experience in certain profes-
sional fields. The knowledge is presented in the format of
IF–THEN and stored in the knowledge base. The mecha-
nism processing the knowledge is called the inference
engine. It is employed to produce a conclusion for some
conditions. An expert system primarily consists of a knowl-
edge base, an inference engine, and an interface. The CRIS
0 5 10 20 50 200 cc
was designed by integrating the three components.
very few few fair much v ery much
The knowledge base stores the facts and rules in two cat-
Fig. 6. Membership functions for oil leakage. egories: machine faults and general customer requirements.
The presentation of knowledge can be one, or hybrid, of
the following: Rule Base, Semantic Network, Object-Attri-
bute-Value, Frame Base and Logic Base. The approach of
predicate level is provided in Fig. 6. In this example, ‘‘very
this investigation uses IF–THEN rule to present the knowl-
few’’ is an X-shaped function, ‘‘very much’’ an S-shape,
edge, forming a knowledge base. Because the rule validity
and P-shape for the others.
is uncertain, or the problem statement of a customer might
The requirements were received from a customer and
be ambiguous, or the problem itself has many possible
stated as ‘‘We have a problem of the oil pump. It leaks seri-
solutions, the uncertainties become more difficult to deal
ously and is not working. The temperature of the pump is
with problems. An inference engine was therefore con-
as high as 280 °F. The leakage is about 60 c.c.’’ The system
structed to ascertain the root causes and store them into
will fuzzify the input by searching the fuzzy proposition
the knowledge base.
base and determining two predicates as the following:
The expert system contains 158 rules in total. The exam-
ples are as follows:
1. Pump temperature, x denoted in Fahrenheit, is high:
(
x200 Rule 11:
150
; 200 < x 6 350
T ¼ 500x IF machine type is VS-50
150
; 350 < x 6 500
and injection failure
2. Volume of oil leakage, y denoted in c.c., is much: and oil temperature is high
( y20 and mold is malfunctioning
; 20 < x 6 50
30
L ¼ 200y and object intrusion failure
150
; 50 < x 6 200 and inlet pipe not heated up to settings
THEN outer steel pipe is broken or loosened
The system then searched the fuzzy rule base to retriev-
ing related rules. Two rules were found as follows: Rule 22:
IF machine type is VS-90
1. Rule 112: IF pump temperature is high AND oil leakage and loud noise
is much THEN oil pump is broken. and radial vibration of motor support fails to reduce
2. Rule 213: IF oil leakage is much AND motor noise is high and injection precision has decreased
THEN parking ring is damaged OR pipe screw is loose. and machine variation has increased
THEN ball screw is curved
Rule 112 contains two predicates of the requirement
statement, much and high. The implication computation The function of the inference engine is to use various
of H and L was then performed in the fuzzy strategy base: inference algorithms to propose root causes by comparing
with facts and rules in the knowledge base. The inference
Z
process will produce new facts and rules, which will be
T  L ¼ Rðx; yÞ ¼ lT ðxÞ ^ lL ðyÞ=ðx; yÞ
stored in the knowledge base. As the knowledge base accu-
mulates, the uncertainty difficulty can be minimized.
yielding product of 0.86667. The process of coordinating inference engine and knowl-
Rule 213 contains one predicate of the requirements edge base is illustrated with an example shown below.
statement. The computational result of membership is 0. The following problems for a machine, type VS-90,
The THEN operator performs a mapping-like function. occurred: (1) mold is not easily aligned; (2) injection parts
After comparing those two rules, rule 112 was adopted to spread on the floor; (3) dimensions of injection parts is
infer the root cause of the problem and achieve the result apparently not identical; (4) radial vibration of motor sup-
of ‘‘oil pump is broken.’’ ports cannot be reduced and (5) machine is operating with
Y.-S. Juang et al. / Expert Systems with Applications 32 (2007) 868–878 875

loud noise. The operator, a user at the customer site, or the Table 3
service personnel at the manufacturing site, enter those set- Fact–Rule inference processes
tings into the CRIS by selecting appropriate predicates Facts Rules
under an appropriate terminology. The system searches I,II,III,IV,VIII 5 I,II ! V
the rule base, retrieving four rules (rule 5, 14, 22 and 37). I,II,III,IV,V,VIII 37 V,VIII ! VII
Through the forward-inference mechanism and with the I,II,III,IV,V,VII,VIII 14 III ! VI
I,II,III,IV,V,VI,VII, VIII 22 IV,VI,VII,VIII ! Root cause
help of the knowledge base, the inference engine will fur-
ther identify the root causes. The inference process is
depicted in the following:
Root cause inferred: ball screw is curved.
1. Fact I: the mold is not easily aligned
Fact II: injection parts spread on the floor Tables 1–3 summarize the inference processes.
Rule 5: IF mold is not easily aligned AND injection
parts spread on the floor THEN machine base is 4.6. System architecture
vibrating.
Fact inferred (Fact V): machine base is vibrating. The CRIS performs many activities to handle customer
2. Fact VIII: loud noise requirement, such as information collection, classification,
Fact V: machine base is vibrating fuzzification, rule evaluation, inference, aggregation and
Rule 37: IF loud noise AND machine base vibrating defuzzification as depicted in Fig. 7. The processes result
THEN machine is vibrating dramatically. in crisp target values for handling customer requirement
Fact inferred (Fact VII): machine base is vibrating or root causes of machine faults. When requirement infor-
dramatically. mation has been collected and the customer information
3. Fact III: dimension of injection parts are apparently gathered, the system activates the classification module to
not identical group the request. When a set of customer requirements
Rule 14: IF dimension of injection parts appears not has been identified as a machine-related requirement, the
to be identical THEN operation fails to maintain attributes are first treated according to the term sets (sets
precision. of linguistic variables) of the relevant terminology and
Fact inferred (Fact VI): low operation precision. transformed by the fuzzy set hedges as appropriate to fully
4. Fact VIII: loud noise elaborate the maintenance attributes, as has been depicted
Fact IV: dimension of injection parts apparently not in the previous section in this study. This process was
identical applied to all the ambiguously described customer require-
Fact VI: low injection precision ments. Processes of expert systems were also triggered to
Fact VII: machine vibrating dramatically diagnose the machine faults with the aid of the knowledge
Rule 22: IF loud noise AND radial vibration of base and inference engine. Using the inferred fact, an intact
motor supports not reduced AND low injection pre- fuzzy rule base was developed, to capture the customer
cision AND machine vibrating dramatically THEN requirement knowledge and experience, defining the rela-
ball screw is curved. tionship between customer requirements and machine diag-
nosis characteristics. The knowledge base for machine fault
Table 1
diagnosis was constructed by domain experts and was
Facts summary
updated by the system. The system is capable of processing
Facts
customer requirements and providing proper recommenda-
I Mold not easily aligned tions to the specific requirement. The preset diagnosis mod-
II Injection parts spread on the floor
ule and recommendation database make this possible.
III Dimension of injection parts apparently not identical
IV Radial vibration of motor supports not reduced
V Machine base vibrating 4.7. System implementation
VI Low injection precision
VII Machine vibrating dramatically The prototype system was developed by many software
VIII Loud noise
packages: Access, Visual Basic, and Active Server Pages

Table 2
Rules summary
Rules
5 IF mold is not easily aligned AND injection parts spread on the floor THEN machine base is vibrating
37 IF loud noise AND machine base vibrating THEN machine is vibrating dramatically
14 IF dimension of injection parts appears not to be identical THEN operation fails to maintain precision
22 IF loud noise AND radial vibration of motor supports not reduced AND low injection precision AND machine
vibrating dramatically THEN ball screw is curved
876 Y.-S. Juang et al. / Expert Systems with Applications 32 (2007) 868–878

Inexact Customer Machine/


Requirement Maintenance
Information Information

Collection and Mapping and Specification


Classification Query and Analysis

KCR KCRMA KMA


(MF) (FIR) (ER)

Rule Evaluation & Inference

Customer
User Interface Requirement &
Machine/
Maintenance
Attribute

KCR: Knowledge about Customer Requirement


KMA: Knowledge about Machine Attributes
KCRMA: Knowledge about the relationship between Customer Requirement and Machine Attributes
MF: Membership Functions
ER: Expert System Rules
FIR: Fuzzy Inference Rules/Propositions

Fig. 7. Architecture of fuzzy inference system.

(ASP). Access helped design the database. Visual Basic was commands from remote sites, such as sales agents, mainte-
used to construct the user interface, while ASP facilitated nance centers or authorized end customers. It also
connection and modification of the system database, inte- responds to the managers of the manufacturers while they
grating the whole system. The prototype of CRIS is run- perform inquiries such as machine performance analysis or
ning on a web server in a local machine tool statistical analysis of customer requirement.
manufacturer. After test activity and system refinement,
the performance of the system is promising. 4.7.2. Design of database
The design of database for the prototype system is
4.7.1. System operation configuration mainly based on relational database with Access. The set-
The CRIS is configured for installation at the manufac- ting of primary keys and database normalization are inev-
turing site. It is installed in a web server running database itable for implementing a database application system
applications. The system users are primarily customers and successfully.
maintenance personnel of the machine tool manufacturer. The primary key is a field that uniquely describes each
The CRIS database is updated while the system receives record. In the Requirement Classification System, for
Y.-S. Juang et al. / Expert Systems with Applications 32 (2007) 868–878 877

Fig. 8. Design of system interface.

instance, the customer ID number is set to be the primary poses and develops an information system for customer
key in the data table storing customer information. The requirement handling in the machine tool industry.
content of this field is unique and generated by the system Fuzzy inference and expert systems are utilized in this
automatically. There are seven major tables used in the study. When a customer requirement is encountered, fuzzy
CRIS database including customer table, requirement inference is adopted to articulate the unclear parts of
table, personnel table, components table, fuzzy rule base requirement wordings. CRIS employs expert systems to
related tables, knowledgebase related tables, and expert infer the root causes of problems and to suggest an appro-
system related tables. Primary keys are associated with a priate solution to those problems.
field in other tables according to a specific relation. Furthermore, this system also helps the customer service
department to analyze customer feedback and integrate it
4.7.3. Design of interface with after-sale service records. By analyzing the service
The system interface was designed on the basis of system records, product quality can be improved and the service
usability. It allows users to input data by clicking on a tag can be enhanced.
and making their choice directly from the list. Fig. 8 pre- Recent advances in information technology, companies
sents a display image for fuzzy handling of a customer have provided an opportunity for significant improvement
requirement. in customer requirement handling. However, several
important issues still need to be adequately addressed by
5. Conclusion future work. The system developed in this study also needs
more refinement. Further work will direct to enhancing
Requirement handling activities play an important role inference capacity of the expert system and performing
in after-sale service within the machine tool industry. more functional analysis from the database.
After-sale service is a critical factor in maintaining cus-
tomer royalty. During recent years, the development of Acknowledgement
information systems has assisted the companies in solving
many managerial problems, to keep business running effi- The authors would like to thank the National Science
ciently. It is useful and valuable to use information systems Council, Taiwan, for partially supporting this research
to handle customer requirement problems. This study pro- under Contract No. NSC90-2218-E-167-004.
878 Y.-S. Juang et al. / Expert Systems with Applications 32 (2007) 868–878

References Lin, P.-S. (1998). Applying expert systems in fault diagnosis. Unpublished
Master Thesis, National Taiwan University.
Berkan, R. C., & Trubatch, S. L. (1997). Fuzzy systems design principles— Liu, S.-Y., & Chen, J.-G. (1995). Development of a machine trouble-
Building fuzzy if–then rules bases. New York: IEEE Press. shooting expert system via fuzzy multi-attribute decision-making
Bravoco, R., & Yadav, S. (1985). Methodology to model functional approach. Expert Systems with Applications, 8(1), 187–201.
structure of an organization. Computers in Industry, 6, 345–361. Mayor, R. J., Benjamin, P. C., Bruce, E., & Painter, C. M. (1995). A
Chen, C.-Y. (2001). Framework and circuitry design for a fuzzy logic con- framework and a suit of methods for business process reengineering. In
troller. Unpublished Master Thesis, National Cheng-Kung University. V. Grover & W. J. Kettinger (Eds.), Business Process Change:
Chen, G., & Pham, T. T. (2000). Introduction to fuzzy sets, fuzzy logic, and Reengineering Concept, Method and Technologies. Idea Group
fuzzy control systems. Boca Raton, FL: CRC Press. Publishing.
Chi, Y.-C., & Zhang, X.-D. (1999). Fuzzy set theory and applications. Ou, J. (1997). The perspectives of machining tool industry in Taiwan
Taipei: Ruling Publishing Co. [Special issue for Machining Tool Industry] (9th ed.). Taipei: Economy
Cristo, B. (1997). An experimental study of service recovery options. Daily.
International Journal of Service Industry Management, 8, 110–130. Perng, C., Li, W-C., Tsai, J.-T., & Lin, S.-S. (1996). A machine fault
Day, R. L. (1980). Research perspectives on consumer complaining diagnosis system using neural network. In Proceedings of the 1996
behavior. Theoretical Development in Marketing, 211–215. annual conference for Chinese Institute of Industrial Engineer.
Fang, X. D. (1995). Expert system-supported fuzzy diagnosis of finish- Perng, C., Tsai, J.-T., & Lin, S.-S. (2003). A design of collaborative
turning process states. International Journal of Machine Tools and commerce service system: Taiwan machine tool industries as an
Manufacture, 35(6), 913–924. example. In Proceedings of the Portland international conference on
Gronroos, C. (1988). Service quality: The six criteria of good perceived management of engineering & technology (PICMET), Portland, OR,
service quality. Review of Business, 9, 10–13. USA.
Hart, W. L., Heskett, J. L., & Sasser, W. E. (1990). The profitable art of Ross, D. T. (1985). Application of SADT. IEEE Transactions on
service recovery. Harvard Business Review, 68, 148–156. Computers(April), 25–34.
Heskett, J. L., Jones, T. O., Loveman, G. W., Sasser, W. E., & Schlesinger, Ruan, D., & Kerre, E. E. (2000). Fuzzy if–then rules in computational
L. A. (1994). Putting the service-profit chain to work. Harvard Business intelligence—Theory and applications. Dordrecht, MA: Kluwer Aca-
Review, 72, 164–174. demic Publishers.
Jacoby, J., & Jaccard, J. J. (1981). The source, meaning and validity of Su, G.-W. (2000). Development and application of fault recovery
consumer complaining behavior: A psychological review. Journal of mechanism based on knowledge management. Unpublished Master
Retailing, 57, 4–24. Thesis, National Chinghua University.
Javadpour, R., & Knapp, G. M. (2003). A fuzzy neural network approach Tsai, J.-T., Shuei, J.-C., Perng, C., Lin, S.-S., & Yao, M.-C. (2001). An
to machine condition monitoring. Computers and Industrial Engineer- information system of maintenance component management for
ing, 45, 323–330. machine tool industry. In Proceedings of the 2001 annual conference
Joentgen, A., Mikenina, L., Weber, R., Zeugner, A., & Zimmermann, H. for Chinese Institute of Industrial Engineer.
J. (1999). Automatic fault detection in gearboxes by dynamic fuzzy Tsai, J.-T., Lu, C.-C., Perng, C., Lin, S.-S., & Chu, H.-C. (2001). A
data analysis. Fuzzy Sets and Systems, 105, 123–132. structure of the marketing executive information system for the
Kalakota, R., & Robinson, M. (1999). E-business roadmap for success. machine tool industry. In Proceedings of the 2001 annual conference of
New Jersey: Addison-Wesley. Chinese Institute of Industrial Engineers.
Lewis, R. C. (1982). Consumers complaint: What happens when business Zadeh, L. A. (1975). The concept of a linguistic variable and its
responds? Proceedings of Consumer Satisfaction/Dissatisfaction and applications to approximate reasoning, part 1, 2 and 3. Information
Complaining Behavior Conference, 88–94. Science, 8, 199–249, 8, 301–357; 9, 43–80.